Artificial
Intelligence
—
Technical
Skills
01
Programming
&
Software
Engineering
02
Mathematics
&
Statistics
03
Machine
Learning
Foundations
04
Deep
Learning
05
Natural
Language
Processing
06
Large
Language
Models
&
Generative
AI
07
Computer
Vision
08
Reinforcement
Learning
09
Data
Engineering
&
Pipelines
10
MLOps
&
Production
Systems
11
Cloud
,
Infrastructure
&
Accelerators
12
Responsible
AI
,
Security
&
Governance
13
Research
,
Tooling
&
Communication
1 / 14
01
Programming
&
Software
Engineering
The
engineering
bedrock
.
Every
AI
system
is
software
first
—
readable
,
testable
,
versioned
,
and
maintainable
code
is
a
prerequisite
,
not
an
afterthought
.
Python
Primary
language
for
ML
—
idiomatic
syntax
,
type
hints
,
virtual
environments
,
packaging
with
pyproject
/
uv
,
async
I
/
O
.
C
++ &
CUDA
C
Performance
-
critical
kernels
,
custom
operators
,
CUDA
extensions
for
PyTorch
,
memory
-
bound
optimization
.
SQL
Analytical
querying
,
window
functions
,
CTEs
,
query
plan
reading
,
warehouse
-
specific
dialects
.
Bash
&
Shell
Scripting
Job
orchestration
,
environment
setup
,
log
parsing
,
automation
of
training
and
evaluation
pipelines
.
Git
&
Version
Control
Workflows
Branching
strategies
,
rebasing
,
code
review
,
submodules
,
hooks
,
monorepo
tooling
with
Git
LFS
for
model
artifacts
.
Software
Design
&
Testing
Modular
architecture
,
separation
of
concerns
,
pytest
and
property
-
based
testing
,
mocking
,
dependency
injection
.
Performance
Profiling
cProfile
,
py
-
spy
,
memory
profilers
,
flame
graphs
,
identifying
CPU
–
GPU
bottlenecks
.
Compiled
Acceleration
Numba
JIT
,
Cython
,
C
++
bindings
via
pybind
11,
Rust
extensions
for
latency
-
sensitive
services
.
API
Design
REST
and
gRPC
service
contracts
,
versioning
,
pagination
,
idempotency
,
request
validation
.
Code
Review
&
Documentation
Technical
writing
,
docstrings
,
ADRs
,
reproducible
notebook
-
to
-
library
hygiene
.
2 / 14
02
Mathematics
&
Statistics
The
theoretical
substrate
.
Model
behavior
,
failure
modes
,
and
optimization
limits
are
all
explained
here
before
they
appear
in
code
.
Linear
Algebra
Matrix
decompositions
,
eigendecomposition
,
SVD
,
tensor
operations
,
low
-
rank
approximation
.
Probability
Theory
Distributions
,
conditional
probability
,
Bayes
'
theorem
,
expectation
,
concentration
inequalities
.
Statistical
Inference
Hypothesis
testing
,
confidence
intervals
,
bootstrapping
,
power
analysis
,
multiple
-
comparison
correction
.
Experimental
Design
A
/
B
and
switchback
testing
,
randomization
units
,
sequential
testing
,
guardrail
metrics
,
novelty
effects
.
Calculus
&
Optimization
Theory
Gradients
,
Jacobians
,
Hessians
,
convexity
,
Lagrangian
duality
,
constrained
optimization
.
Information
Theory
Entropy
,
cross
-
entropy
,
KL
divergence
,
mutual
information
,
rate
–
distortion
tradeoffs
.
Numerical
Methods
Floating
-
point
precision
,
numerical
stability
,
conditioning
,
mixed
-
precision
arithmetic
.
Graph
&
Discrete
Mathematics
Graph
theory
,
combinatorics
,
discrete
structures
underlying
graph
neural
networks
and
search
.
Causal
Inference
DAGs
,
do
-
calculus
,
instrumental
variables
,
difference
-
in
-
differences
,
propensity
scoring
.
Time
-
Series
Analysis
Stationarity
,
autocorrelation
,
spectral
methods
,
state
-
space
models
,
seasonality
decomposition
.
3 / 14
03
Machine
Learning
Foundations
Classical
algorithms
and
the
evaluation
discipline
that
surrounds
them
.
Most
production
wins
still
come
from
well
-
tuned
classical
models
.
Supervised
Learning
Linear
and
logistic
regression
,
SVMs
,
decision
trees
,
k
-
NN
,
naive
Bayes
.
Unsupervised
Learning
k
-
means
,
DBSCAN
,
hierarchical
clustering
,
PCA
,
t
-
SNE
,
UMAP
,
density
estimation
.
Ensemble
Methods
Random
forests
,
gradient
boosting
with
XGBoost
,
LightGBM
,
CatBoost
,
stacking
and
blending
.
Feature
Engineering
Encoding
,
binning
,
interactions
,
target
encoding
,
embeddings
as
features
,
feature
leakage
detection
.
Model
Evaluation
Cross
-
validation
,
stratified
splits
,
ROC
-
AUC
,
PR
curves
,
calibration
,
confusion
analysis
.
Regularization
&
Tuning
L
1/
L
2,
early
stopping
,
grid
,
random
,
and
Bayesian
search
with
Optuna
or
Ray
Tune
.
Imbalanced
Learning
Resampling
,
class
weighting
,
focal
loss
,
threshold
optimization
against
business
cost
functions
.
Anomaly
Detection
Isolation
forests
,
one
-
class
SVM
,
autoencoder
reconstruction
error
,
statistical
process
control
.
Recommender
Systems
Collaborative
filtering
,
matrix
factorization
,
two
-
tower
retrieval
,
ranking
and
re
-
ranking
.
Probabilistic
Modeling
Bayesian
inference
,
MCMC
,
variational
methods
,
Gaussian
processes
,
probabilistic
graphical
models
.
4 / 14
04
Deep
Learning
Architectures
,
training
dynamics
,
and
the
systems
engineering
required
to
make
large
models
converge
reliably
.
Neural
Network
Fundamentals
Backpropagation
,
activation
functions
,
weight
initialization
,
normalization
layers
,
residual
connections
.
Frameworks
PyTorch
,
TensorFlow
/
Keras
,
JAX
with
Flax
or
Equinox
,
Lightning
for
training
scaffolding
.
Core
Architectures
CNNs
,
RNNs
and
LSTMs
,
transformers
,
U
-
Nets
,
GANs
,
diffusion
models
,
mixture
-
of
-
experts
.
Training
Optimization
AdamW
,
learning
-
rate
schedules
,
warmup
,
gradient
clipping
,
loss
scaling
,
curriculum
learning
.
Distributed
Training
Data
,
tensor
,
and
pipeline
parallelism
;
DDP
,
FSDP
,
DeepSpeed
,
Megatron
-
LM
,
ZeRO
stages
.
Mixed
Precision
&
Memory
FP
16
and
BF
16
training
,
gradient
accumulation
,
activation
checkpointing
,
offloading
strategies
.
Regularization
Techniques
Dropout
,
weight
decay
,
label
smoothing
,
stochastic
depth
,
data
augmentation
pipelines
.
Transfer
Learning
Pretraining
and
fine
-
tuning
,
frozen
backbones
,
layer
-
wise
learning
rates
,
domain
adaptation
.
Model
Compression
Pruning
,
knowledge
distillation
,
structured
sparsity
,
quantization
-
aware
training
.
Debugging
&
Profiling
Autograd
inspection
,
gradient
checking
,
torch
.
profiler
,
Nsight
Systems
,
loss
-
curve
forensics
.
5 / 14
05
Natural
Language
Processing
Text
and
speech
pipelines
:
tokenization
,
representation
,
task
-
specific
modeling
,
and
the
metrics
that
decide
whether
any
of
it
worked
.
Tokenization
Byte
-
pair
encoding
,
WordPiece
,
SentencePiece
,
subword
regularization
,
vocabulary
design
.
Text
Representation
Word
2
Vec
,
GloVe
,
fastText
,
contextual
embeddings
,
sentence
-
transformers
.
Sequence
Labeling
Named
entity
recognition
,
part
-
of
-
speech
tagging
,
slot
filling
,
span
extraction
.
Text
Classification
Sentiment
,
intent
detection
,
topic
tagging
,
multi
-
label
and
hierarchical
classification
.
Machine
Translation
Encoder
–
decoder
models
,
multilingual
transfer
,
back
-
translation
,
terminology
constraints
.
Summarization
Extractive
and
abstractive
approaches
,
long
-
document
handling
,
faithfulness
evaluation
.
Information
Retrieval
BM
25,
dense
retrieval
,
hybrid
search
,
cross
-
encoder
re
-
ranking
,
chunking
strategy
.
Question
Answering
Extractive
and
generative
QA
,
open
-
domain
retrieval
,
multi
-
hop
reasoning
.
Speech
Processing
Automatic
speech
recognition
with
Whisper
,
text
-
to
-
speech
synthesis
,
speaker
diarization
.
NLP
Evaluation
BLEU
,
ROUGE
,
METEOR
,
BERTScore
,
perplexity
,
human
preference
studies
.
6 / 14
06
Large
Language
Models
&
Generative
AI
The
fastest
-
moving
layer
of
the
stack
—
pretraining
,
adaptation
,
alignment
,
retrieval
,
agents
,
and
the
economics
of
serving
them
.
Transformer
Internals
Multi
-
head
attention
,
rotary
and
ALiBi
position
encodings
,
KV
caching
,
grouped
-
query
attention
.
Pretraining
Corpus
curation
and
deduplication
,
data
mixing
,
scaling
laws
,
Chinchilla
-
optimal
token
budgets
.
Fine
-
Tuning
Supervised
fine
-
tuning
,
LoRA
and
QLoRA
,
adapters
,
prefix
and
prompt
tuning
,
instruction
datasets
.
Alignment
RLHF
,
DPO
and
IPO
,
reward
modeling
,
constitutional
and
RLAIF
methods
,
refusal
calibration
.
Prompt
Engineering
Few
-
shot
design
,
chain
-
of
-
thought
,
decomposition
,
structured
output
via
JSON
schema
,
prompt
versioning
.
Retrieval
-
Augmented
Generation
Embedding
selection
,
vector
stores
,
hybrid
retrieval
,
re
-
ranking
,
context
window
budgeting
,
citation
grounding
.
Agentic
Systems
Tool
and
function
calling
,
planning
and
decomposition
,
memory
architectures
,
multi
-
agent
orchestration
.
Inference
Optimization
Quantization
,
PagedAttention
,
continuous
batching
,
speculative
decoding
,
vLLM
,
TensorRT
-
LLM
.
Generative
Media
Latent
diffusion
,
ControlNet
conditioning
,
inpainting
,
text
-
to
-
image
and
text
-
to
-
video
pipelines
.
LLM
Evaluation
&
Guardrails
LLM
-
as
-
judge
,
hallucination
detection
,
red
-
teaming
,
jailbreak
resistance
,
toxicity
filtering
.
7 / 14
07
Computer
Vision
Recognition
,
detection
,
segmentation
,
and
3
D
reconstruction
—
with
the
annotation
infrastructure
that
vision
work
quietly
depends
on
.
Image
Classification
ResNet
,
EfficientNet
,
ConvNeXt
,
Vision
Transformers
,
Swin
,
transfer
from
large
pretrained
backbones
.
Object
Detection
YOLO
family
,
Faster
R
-
CNN
,
DETR
and
deformable
DETR
,
anchor
-
free
detectors
,
NMS
tuning
.
Semantic
&
Instance
Segmentation
Mask
R
-
CNN
,
U
-
Net
,
DeepLab
,
Segment
Anything
,
panoptic
segmentation
.
Multimodal
Vision
-
Language
CLIP
,
SigLIP
,
vision
-
language
models
,
image
captioning
,
visual
question
answering
.
Video
Understanding
Temporal
modeling
,
action
recognition
,
multi
-
object
tracking
,
optical
flow
,
video
summarization
.
3
D
Vision
Point
clouds
with
PointNet
++,
depth
estimation
,
NeRF
,
Gaussian
splatting
,
SLAM
.
Document
&
OCR
Layout
analysis
,
table
extraction
,
handwritten
text
recognition
,
structured
document
parsing
.
Generative
Vision
Diffusion
models
,
GANs
,
super
-
resolution
,
image
editing
,
style
transfer
.
Data
Augmentation
Geometric
transforms
,
MixUp
,
CutMix
,
RandAugment
,
synthetic
data
generation
.
Annotation
Tooling
Label
Studio
,
CVAT
,
Scale
,
labeling
guidelines
,
inter
-
annotator
agreement
measurement
.
8 / 14
08
Reinforcement
Learning
Sequential
decision
-
making
under
uncertainty
—
from
classic
control
to
the
alignment
pipelines
that
shape
modern
language
models
.
Markov
Decision
Processes
States
,
actions
,
rewards
,
discounting
,
Bellman
equations
,
policy
and
value
iteration
.
Value
-
Based
Methods
Q
-
learning
,
SARSA
,
DQN
,
double
and
dueling
DQN
,
prioritized
experience
replay
.
Policy
Gradient
Methods
REINFORCE
,
A
2
C
,
A
3
C
,
PPO
,
TRPO
,
advantage
estimation
with
GAE
.
Continuous
Control
DDPG
,
TD
3,
SAC
,
model
predictive
control
,
trajectory
optimization
.
Model
-
Based
RL
Learned
dynamics
models
,
Dreamer
,
MCTS
,
planning
with
rollouts
.
Multi
-
Agent
RL
Cooperative
and
competitive
settings
,
self
-
play
,
emergent
behavior
,
credit
assignment
.
Reward
Engineering
Reward
shaping
,
sparse
-
reward
handling
,
inverse
RL
,
reward
model
training
.
Simulation
&
Sim
-
to
-
Real
Domain
randomization
,
Isaac
Gym
,
MuJoCo
,
PyBullet
,
reality
-
gap
mitigation
.
Bandits
&
Exploration
Thompson
sampling
,
UCB
,
contextual
bandits
,
off
-
policy
evaluation
.
9 / 14
09
Data
Engineering
&
Pipelines
Models
are
downstream
of
data
.
Ingestion
,
transformation
,
versioning
,
and
quality
control
determine
the
ceiling
on
model
performance
.
Data
Wrangling
pandas
,
Polars
,
NumPy
,
Arrow
,
efficient
joins
,
reshaping
,
memory
-
aware
processing
.
Distributed
Processing
Apache
Spark
,
Ray
,
Dask
,
distributed
DataFrame
patterns
,
shuffle
and
skew
tuning
.
Streaming
Data
Kafka
,
Flink
,
Kinesis
,
event
-
time
processing
,
exactly
-
once
semantics
,
windowing
.
Pipeline
Orchestration
Airflow
,
Dagster
,
Prefect
,
dependency
graphs
,
retries
,
backfills
,
SLA
alerting
.
Data
Modeling
&
Warehousing
Dimensional
modeling
,
dbt
,
Snowflake
,
BigQuery
,
Databricks
,
medallion
architecture
.
Feature
Stores
Feast
,
Tecton
,
online
–
offline
consistency
,
point
-
in
-
time
correctness
,
feature
reuse
.
Data
Quality
Great
Expectations
,
schema
validation
,
freshness
checks
,
anomaly
alerting
,
contract
testing
.
Data
Versioning
&
Lineage
DVC
,
LakeFS
,
Delta
Lake
time
travel
,
column
-
level
lineage
,
provenance
tracking
.
Annotation
&
Labeling
Ops
Workforce
design
,
gold
-
standard
sets
,
active
learning
for
label
efficiency
,
quality
audits
.
Privacy
-
Preserving
Data
Handling
PII
detection
and
redaction
,
synthetic
data
generation
,
de
-
identification
,
consent
tracking
.
10 / 14
10
MLOps
&
Production
Systems
Where
models
meet
users
.
Deployment
,
observability
,
and
the
operational
habits
that
keep
systems
trustworthy
after
launch
.
Experiment
Tracking
MLflow
,
Weights
&
Biases
,
Neptune
,
run
comparison
,
artifact
logging
,
reproducibility
metadata
.
Model
Packaging
&
Registry
ONNX
,
TorchScript
,
SavedModel
,
model
cards
,
stage
promotion
,
artifact
signing
.
CI
/
CD
for
ML
Automated
testing
,
data
validation
gates
,
shadow
deployments
,
canary
rollouts
,
blue
-
green
releases
.
Model
Serving
FastAPI
,
TorchServe
,
Triton
Inference
Server
,
KServe
,
BentoML
,
streaming
and
batch
endpoints
.
Monitoring
&
Observability
Data
drift
and
concept
drift
detection
,
prediction
distribution
shifts
,
latency
SLOs
,
alert
design
.
Incident
Response
Rollback
procedures
,
degraded
-
mode
fallbacks
,
root
-
cause
analysis
,
postmortem
practice
.
Scale
&
Cost
Engineering
Batching
and
concurrency
tuning
,
autoscaling
policies
,
GPU
utilization
,
cost
-
per
-
inference
accounting
.
Reproducibility
Seed
management
,
containerized
environments
,
dependency
pinning
,
deterministic
training
runs
.
Evaluation
Harnesses
Offline
benchmark
suites
,
human
-
in
-
the
-
loop
review
,
online
A
/
B
validation
,
regression
test
sets
.
Platform
Engineering
Internal
ML
platforms
,
self
-
service
training
,
notebook
-
to
-
production
paths
,
developer
experience
.
11 / 14
11
Cloud
,
Infrastructure
&
Accelerators
Compute
strategy
determines
what
is
economically
feasible
.
Cloud
platform
fluency
and
accelerator
literacy
are
now
core
AI
skills
.
AWS
SageMaker
,
S
3,
EC
2
GPU
instances
,
Bedrock
,
EKS
,
Step
Functions
,
IAM
policy
design
.
Google
Cloud
Vertex
AI
,
BigQuery
,
Cloud
TPUs
,
GKE
,
Dataflow
,
Artifact
Registry
.
Microsoft
Azure
Azure
Machine
Learning
,
Azure
OpenAI
Service
,
AKS
,
Synapse
,
managed
identity
.
Containerization
Docker
,
multi
-
stage
builds
,
image
slimming
,
CUDA
base
images
,
reproducible
environments
.
Kubernetes
Deployments
,
autoscaling
,
GPU
scheduling
with
device
plugins
,
Helm
,
operators
,
node
pools
.
GPU
Programming
CUDA
,
cuDNN
,
NCCL
,
Triton
kernels
,
shared
memory
,
occupancy
tuning
,
tensor
cores
.
Accelerator
Landscape
A
100
and
H
100
class
GPUs
,
TPUs
,
inference
accelerators
,
interconnect
bandwidth
tradeoffs
.
Infrastructure
as
Code
Terraform
,
Pulumi
,
CloudFormation
,
environment
parity
,
drift
detection
.
Storage
&
Networking
Object
storage
patterns
,
dataset
sharding
,
high
-
throughput
data
loading
,
VPC
and
egress
cost
control
.
Cluster
Orchestration
SLURM
,
Ray
clusters
,
job
queues
,
spot
and
preemptible
instance
strategies
.
12 / 14
12
Responsible
AI
,
Security
&
Governance
Increasingly
a
gating
requirement
rather
than
an
optional
specialization
.
Fairness
,
explainability
,
privacy
,
and
adversarial
robustness
.
Fairness
&
Bias
Mitigation
Demographic
parity
,
equalized
odds
,
disparate
impact
analysis
,
pre
-,
in
-,
and
post
-
processing
remedies
.
Explainability
SHAP
,
LIME
,
permutation
importance
,
counterfactual
explanations
,
attention
and
saliency
analysis
.
Privacy
Engineering
Differential
privacy
,
federated
learning
,
secure
aggregation
,
PII
redaction
,
data
minimization
.
Adversarial
Robustness
Evasion
attacks
,
poisoning
,
model
extraction
,
prompt
injection
defense
,
adversarial
training
.
AI
Governance
&
Regulation
EU
AI
Act
risk
tiers
,
GDPR
,
NIST
AI
RMF
,
ISO
/
IEC
42001,
model
documentation
requirements
.
Model
Documentation
Model
cards
,
datasheets
for
datasets
,
intended
-
use
statements
,
limitation
disclosure
.
Evaluation
for
Safety
Red
-
teaming
methodologies
,
harm
taxonomies
,
stress
testing
,
safety
benchmark
design
.
Human
-
in
-
the
-
Loop
Design
Escalation
paths
,
confidence
thresholds
,
review
interfaces
,
automation
bias
mitigation
.
Auditability
Immutable
logs
,
decision
traceability
,
versioned
prompts
and
policies
,
audit
-
ready
evidence
trails
.
13 / 14
13
Research
,
Tooling
&
Communication
The
multiplier
skills
.
Research
literacy
and
clear
communication
determine
how
quickly
good
ideas
reach
production
.
Research
Literacy
Reading
and
critiquing
papers
,
arXiv
monitoring
,
distinguishing
claims
from
evidence
,
reproduction
studies
.
Applied
Experimentation
Hypothesis
framing
,
ablation
design
,
statistical
significance
,
negative
-
result
documentation
.
Data
Visualization
matplotlib
,
seaborn
,
Plotly
,
TensorBoard
,
Altair
,
chart
selection
and
honest
encoding
.
Technical
Writing
Design
documents
,
RFCs
,
method
sections
,
internal
wikis
,
API
reference
documentation
.
Cross
-
Functional
Collaboration
Working
with
product
,
design
,
legal
,
and
operations
teams
;
translating
model
behavior
into
product
terms
.
Stakeholder
Communication
Executive
briefing
,
uncertainty
communication
,
scoping
expectations
,
risk
framing
.
Mentoring
&
Code
Review
Technical
coaching
,
review
standards
,
onboarding
curriculum
,
knowledge
-
sharing
culture
.
Open
-
Source
Contribution
Upstream
patches
,
issue
triage
,
maintainer
collaboration
,
licensing
awareness
.
Prototyping
Velocity
Rapid
notebook
experimentation
,
hackathon
-
grade
MVPs
,
time
-
boxed
feasibility
probes
.
14 / 14
